AppInsights: PII is logged even if logPersonalInformation is false

Version

4.11.0

Describe the bug

BBF always logs the user’s text.

To Reproduce

1. Have bot with AppInsights integration. Configure the bot to not log the user’s personal information as per doc https://docs.microsoft.com/en-us/azure/bot-service/bot-builder-telemetry?view=azure-bot-service-4.0&tabs=csharp.
2. Talk to the bot
3. Look into any event that logs the Recognizer telemetry, e.g. ValueRecognizerResult
4. Check the customDimensions, you will find “Text” property there. Example:

{
    "conversationId": "0a9d58c0-3301-11eb-bf90-2102e829151f|livechat",
    "activityType": "message",
    "channelId": "emulator",
    "activityId": "2d360170-3301-11eb-98d6-891267dfc0ed",
    "TopIntentScore": "Microsoft.Bot.Builder.IntentScore",
    "Entities": {
        "instance": {
            "Ordinal": [
                {
                    "Name": "Ordinal",
                    "Value": 2,
                    "Start": 4,
                    "End": 9,
                    "Text": "second"
                }
            ]
        }
    },
    "TopIntent": "test2",
    "Intents": {
        "test2": { "score":0.989161491394043 }​
    },
    "Text": "the second test"
}

Expected behavior

User’s text should not be logged if logPersonalInformation == false

Additional context

• The issue is that the LogPersonalInformation property is in TelemetryLoggerMiddleware. It should be a property in IBotTelemetryClient (it should be virtual on BotTelemetryClient).

• FillRecognizerResultTelemetryProperties method should check the property on the logger and it shouldn’t log text or alteredtext if LogPersonalInformation == false.

• Also, we need to do additional analysis to ensure that we don’t log PII in recognizerResult.Properties object.

[Robulane]

@Zerryth, yes, TelemetryLoggerMiddleware handles PII as expected. This bug is about the fact that there other places in BBF that log PII data. Examples that I found:

1. All recognizer results (ValueRecognizerResult, RegexRecognizerResult, RecognizerSetResult, etc.)
2. GeneratorResult can log text that may contain PII data in “text” field

[Zerryth]

Update: still working on this.

I’ve spoken with Eric regarding best practices/design issue that arises around

1. There being a lot of classes that derive from Recognizer
2. Implementing the LogPersonalInformation flag
3. LogPersonalInformation wanting to be a BoolExpression (which needs dependency on AdaptiveExpressions) for adaptive bits, but only a bool for non-adaptive

It goes deeper, however Eric said maybe we should discuss best practices with Gabo, since current solution we can think of isn’t that pretty